Memory system for real agent workflows

MemLayer keeps decisions, constraints, risks, artifacts and reuse paths in one operational memory contour.

It is built for agent loops that need more than a transient prompt buffer: search, relevant context, imports, quality gates, review queues, lifecycle cleanup, compaction and operator visibility.

What MemLayer does

Retrieval

Project-scoped, related-project and global retrieval with lexical, semantic and hybrid modes.

Quality

Quality scoring, semantic duplicate detection, operator review actions and lifecycle maintenance.

Imports

Structured project import, reimport, conflict detection and doc-aware ingestion for real repositories.

Operations

Runtime self-checks, metrics, review queues, task logs, compaction and embedded admin console.

Public entry points

The public surface is intentionally small. Production API lives on a dedicated subdomain. Admin runtime is kept separate and is not linked from this page.

Health check
curl -sS 'https://api.memlayer.ru/health'

Runtime self-check

For a lightweight read-only smoke check there is a dedicated endpoint. Use your own API key and keep it out of shell history if needed.

Read-only self-check
curl -sS 'https://api.memlayer.ru/admin/runtime/self-check?search_query=architecture&limit=5' \
  -H 'Authorization: Bearer YOUR_API_KEY'

Memory quality layer

MemLayer does not only store knowledge. It also checks whether the stored knowledge remains useful: placeholder rejection, semantic duplicate signals, review queues, conflict resolution and compaction are part of the runtime.

Import and reuse

Project scans can bootstrap memory from real repositories, including hidden docs, handoff files and structured artifacts. The system can then retrieve context by project, related projects or globally.

Project search
curl -sS 'https://api.memlayer.ru/memory/search?query=architecture%20decision&mode=hybrid&limit=8' \
  -H 'Authorization: Bearer YOUR_API_KEY'